This code is provided to reproduce experiments in the paper "Layer-stacked Attention for Heterogeneous Graph Embedding".
Running this code requires installation of Python 3.X packages listed in requirements.txt. Install them by running:
pip install -r requirements.txt
It also requires installation of the COGDL Toolkit (version 0.1.2) from the following GitHub repository: https://github.com/THUDM/cogdl To install it, run:
pip install cogdl==0.1.2
To perform either transductive or inductive experiment runs for a method on a dataset, go to the root repository and run the command:
python run.py --method [MetaPath2Vec, HAN, GTN, HGT, LATTE-1, LATTE-2, LATTE-2proximity] --dataset [ACM, DBLP, IMDB] --inductive [True, False] --num_gpus [0, 1]
The wandb logger will generate a link for you inspect the performance metrics as it trains and validates. If prompted, choose the option "(1) Private W&B dashboard, no account required".
The models' hyperparameters have been kept as described in the paper. If you'd like to inspect LATTE's algorithm & code,
open the conv.py file and the LATTENodeClf class in the methods.py file.
Happy Reviewing! --Author